Author Topic: Circulating currents in large arrays of paralleled SMPS's...how to mitigate?  (Read 383 times)

0 Members and 1 Guest are viewing this topic.

Offline treez

  • Super Contributor
  • ***
  • Posts: 1229
  • Country: gb
Hi
We are doing a 12kW power supply by paralleling over 40 DCDC modules. This question is not specifically about the modules. It is rather about the nature of “circulating currents” in such large systems, and how they can be mitigated.

The power supply will be made of ten “blocks” in parallel.  The blocks are shown in the attached.

Obviously in such a physically large system, there will be some distance between  the DCDC modules at either end of the array. Therefore inevitably, we will have large area ground current loops. This is practically unavoidable.

As you know, a current loop, (especially  one enclosing such a large area) is unfortunately a nice “receiving antenna” for noise. So we will end up with large amounts of noise getting coupled into our circuit, and it is highly likely that the power modules will be swamped with noise and will malfunction.

Do you agree with us that our best way to mitigate this noise will be to place “ferrite beads” in all of the current loops, such that any circulating current in any loop….will “see” at least one of the ferrite beads? Also, a common mode choke right at the output of each module.

I am certain we won’t be able to get help from the DCDC module vendor since….
1…Our sales volumes are obviously low
2….We are too insignificant an organisation.
« Last Edit: March 26, 2020, 04:21:02 pm by treez »
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 10643
  • Country: us
  • DavidH
1. There are no ground loops from input to output because the DCM3623T50M53C2T00 are isolated from input to output.  But see 2 below.

2. There is unspecified capacitive coupling between the input and output.  The Vicor datasheet shows an example where input common mode chokes are used to suppress common mode noise.  Common mode chokes are not required on both the input and output.  I would tend to place them on the input like Vicor shows because then it will not affect the load regulation.

3. Pickup or emission depends on the loop area of the high current output wiring which contributes to inductance and antenna aperture.  So the wiring should be arranged like a parallel transmission line to minimize loop area.

4. Another trick that I would probably use is to AC terminate the parallel transmission line at its ends but after the LC filter section shown in Vicor's examples.  This amounts to a sufficiency large capacitor in series with a resistor which matches the impedance of the transmission line.  This is a refinement of the "bulk" decoupling capacitor often recommended except that the ESR is more tightly controlled with a discrete resistor.  This dampens resonances in the necessarily low Q high current wiring.

Adding this is sort of a belt and suspenders approach since there is a good chance it will not be required but I would do it anyway if I lacked the resources to prove it was not required.

5. There are some clever layout arrangements which can equalized the series resistance seen by the regulator outputs.  I am not entirely clear from reading the datasheet how these modules are suppose to share current without a separate connection between them.
« Last Edit: March 26, 2020, 05:27:23 pm by David Hess »
 
The following users thanked this post: treez

Offline treez

  • Super Contributor
  • ***
  • Posts: 1229
  • Country: gb
Thanks,
The modules share current by their  droop algorithm, whereby their Vout steadily drops by 5% from no load to full load.
We are using these modules as non-isolated, and connect primary and secondary grounds directly.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 10643
  • Country: us
  • DavidH
The modules share current by their  droop algorithm, whereby their Vout steadily drops by 5% from no load to full load.

So they have an output resistance commensurate with their voltage accuracy to allow parallel operation.

Quote
We are using these modules as non-isolated, and connect primary and secondary grounds directly.

That complicates the layout.  Ideally the input and output grounds meet at a single point.  That will reduce the effectiveness of the common mode filters at high frequencies so maybe replace them with differential filters.

Can you avoid connecting the input and output grounds together at each converter?
 
The following users thanked this post: treez

Offline filssavi

  • Frequent Contributor
  • **
  • Posts: 300
Ferrite beads are useless in this application, the circulating currents Are usually a low(ish) frequency thing kilohertz at most ( usually much lower) while ferrite beads only start working at in the tens of megahertz range(although you can find a little lower)

What you need is a large inductor (let’s say hundreds of mH at least) at the output of each power supply, of course the higher the switching frequency the lower the inductance needed

The only reason I would see for having such a high number of dc/dc in parallel is if you have an extremely high output current (in the kA range, but in that case I would seriously reconsider your overall design (do like in PC motherboards, a single DC/DC does most of the heavy lifting/ isolation and a multiphase buck that only does the final step, that does not have circulating current issues

 
The following users thanked this post: treez

Offline treez

  • Super Contributor
  • ***
  • Posts: 1229
  • Country: gb
Quote
What you need is a large inductor (let’s say hundreds of mH at least) at the output of each power supply,
Thanks, we cant do that because it would upset the feedback loop.

Quote
That complicates the layout.  Ideally the input and output grounds meet at a single point.  That will reduce the effectiveness of the common mode filters at high frequencies so maybe replace them with differential filters.

Can you avoid connecting the input and output grounds together at each converter?
Thanks, in our case , we are just thinking that its no worse for common mode noise than if we had done Multiple Boost converters in parallel (instead of DCM modules) ...which are also non  isolated.
« Last Edit: March 26, 2020, 09:13:06 pm by treez »
 

Offline duak

  • Frequent Contributor
  • **
  • Posts: 835
  • Country: ca
treez, I have a few questions:

1.) does the input power come in on a single conductor with an earth return or is it actually two conductors in one cable?  My concern is that there is sufficient series inductance that could interact with the negative input resistance of the BCM.  If you don't already know, a switching converter will reduce its current draw when its supply voltage is increased, which is a negative resistance.  This could lead to unexpected voltage variations on the local supply voltage in response to source voltage changes or load current variations.  A load dump where the ouput current drops a significant amount could cause a large spike on the input voltage from the collapsing magnetic field in the inductance.  In addition, there will be some distributed capacitance that will form a transmission line with the series inductance.  The series resistance R1 will dampen the circuit but for something like this, I'd look into the possibility.

2.) why are the input and output circuits tied together in the Power block?  I would have assumed that this would have been done external to the blocks at a higher level.  A common mode choke will help isolate the two circuits over some frequency range but the two circuits will still be connected galvanically.

3). Have you thought about the overal packaging?  I'd sort of expect the power blocks to be flat modules arranged as cards in a card cage (like a Blade server) with power  connections brought to one edge.  Input and output power would be distributed and collected by layered planar conductors.  The output side would be thick copper sheets or bus bars.
 
The following users thanked this post: treez

Offline treez

  • Super Contributor
  • ***
  • Posts: 1229
  • Country: gb
Quote
does the input power come in on a single conductor with an earth return or is it actually two conductors in one cable? 
Just two conductors in one cable. The return conductor may, or may not, end up getting connected to earth on the ground....likely it will,  as otherwise it would be floating.

Quote
does the input power come in on a single conductor with an earth return or is it actually two conductors in one cable?  My concern is that there is sufficient series inductance that could interact with the negative input resistance of the BCM.
Thanks, we have a big electrolytic at the input which will soak up any energy from the cable. It will also mitigate the negative resistance problem.
Quote
why are the input and output circuits tied together in the Power block?  I would have assumed that this would have been done external to the blocks at a higher level.  A common mode choke will help isolate the two circuits over some frequency range but the two circuits will still be connected galvanically.
Thanks,  as you know, all the grounds in the system can’t be floating….they must be tied to something…whether that tie is a direct  connection, or via a 1MEG resistor, the connection has to be  made. So why not just connect together all the grounds everywhere under all the isolated power modules? After all, we don’t need isolation. However, are you implying that  avoiding such direct ground connection would allow a better immunity to noise problems?

Also, if you have the module’s primary and secondary isolated, then you must use Y capacitors to mitigate common mode noise…..whereas direct connection of primary and secondary grounds with copper means you don’t need Y capacitors.

…The problem comes, if you don’t connect input and output ground of each power module, right under the module, …then where do you do it?.....wherever else you pick it wont be as convenient as just connecting module input and output ground locally..right under the module.
« Last Edit: March 28, 2020, 11:36:34 am by treez »
 

Offline duak

  • Frequent Contributor
  • **
  • Posts: 835
  • Country: ca
Treez, I had been following your various posts but missed the one explaining that this is going in a tethered drone.  I thought it might have been something located down a mineshaft with only one conductor available for power.

Are you thinking of circulating currents where a large current builds up because of some sort of resonance?  Vicor mentions this on page 7, 2nd from last paragraph on their paralleling DCMs ap note.  From what I understand about DCMs, their operating frequencies are not synchronized, but can still interact.  I would think that stochastic operation would reduce the possibility of circulating currents.  Very much like solders not marching in step across a bridge.

You also mentioned ground loops.  IMHO, the only way one gets a ground loop is to have a conductive loop in an AC magnetic field that induces an AC current in it.  You can either break the loop or attenuate the AC magnetic field.  Sometimes the problem is bludgeoned to death with a metal chassis with compartments to both confine any fields and provide a low impedance path for any induced currents to flow.  I'd consider breaking the loop by reinstating the isolation between the input & output sides in the power blocks.  The output ground should be connected back to the input gound for safety reasons, ie., to ensure there are no hazards should something fail and leak current from the 700 VDC input to the otherwise floating output circuit.   This connection between the input and output grounds can be made on the chassis and not on the power blocks.  Assuming there's a metal enclosure, to reduce EMI, the input ground should connect to the enclosure where it enters.  The output ground can also connect to the enclosure where it enters.  As to the split power block PCB ground plane, I would think the ground plane under the BCM  would connect to the input ground while the ground plane under the DCM  would connect to the output ground.   There would be a gap in the ground plane between the two sections.

BTW, what evidence do you have for problems from ground loops?  The Vicors I worked with were remarkably clean and didn't have particulary intense near fields.  Have you tried an inductive probe on the parts you have?  If you don't have one, you can make one up from 5 to 10 turns of solid wire wound into a loop of 10 to 20 mm diameter with the ends soldered to a coax cable.  By observing the signal on a 'scope and moving it over the modules you can see where it's coming from, what the prevalent frequencies are and if they're modulated at a particular rate. ie., in bursts.

I hope one of these drones doesn't encounter an energized electrical transmission line.  I'm not sure grounding the 700 V power supply would make much difference.

Cheers,
« Last Edit: Yesterday at 04:46:26 am by duak »
 
The following users thanked this post: treez

Offline treez

  • Super Contributor
  • ***
  • Posts: 1229
  • Country: gb
Quote
BTW, what evidence do you have for problems from ground loops?
Thanks…
1…Somebody  with knowledge of Vicors told us that “circulating currents” would be our biggest problem..they wouldn’t elaborate
2…As you say, we have a big conductive loop in the grounding….so it will have noise induced in to it.
Quote
Are you thinking of circulating currents where a large current builds up because of some sort of resonance?
No.
Quote
Vicor mentions this on page 7, 2nd from last paragraph on their paralleling DCMs ap note. 
Thanks, yes we added the recommended output inductors to mitigate this.
 

Offline treez

  • Super Contributor
  • ***
  • Posts: 1229
  • Country: gb
Quote
I'd consider breaking the loop by reinstating the isolation between the input & output sides in the power blocks.  The output ground should be connected back to the input ground for safety reasons, ie., to ensure there are no hazards should something fail and leak current from the 700 VDC input to the otherwise floating output circuit.   This connection between the input and output grounds can be made on the chassis and not on the power blocks.  Assuming there's a metal enclosure, to reduce EMI, the input ground should connect to the enclosure where it enters.  The output ground can also connect to the enclosure where it enters.  As to the split power block PCB ground plane, I would think the ground plane under the BCM  would connect to the input ground while the ground plane under the DCM  would connect to the output ground.   There would be a gap in the ground plane between the two sections.

Thanks, though I wonder if  the method you describe is better than the attached, where we just have BCM and DCM grounds all connected together,  and just have the  multiple common mode chokes at the input and output of each power block?

The above only involves one grounding connection to the enclosure. Your described method involves at least two connections to the enclosure…of the input and output grounds…this makes a big current loop in itself, so we do not favour this…..or were you thinking of a connection to the enclosure via a 1MEG resistor?

Also, your above described method involves an isolated  ground in between the BCM input and DCM output. Due to this, Y capacitors will need to be connected from the power rails to this “Patch” of isolated ground…whereas if there  was  one contiguous ground  under the BCM’s and DCMs, then  Y capacitors would not be needed, since  the shorting  copper will act as an infinite Y capacitor.
Also, if we have the “patch” of isolated ground that you describe, then we will need to still connect it to the chassis/enclosure, since otherwise it may float up to a high voltage and cause damage to insulation over time. As such we would do this with a 1MEG resistor….but we don’t need to do it at all if we use the attached method with no isolation.

...This again begs the question about  common mode noise in isolated SMPS’s vs NON-Isolated SMPS’s where you only have a two wire connection…..its always easier to counteract  common mode noise in non-isolated SMPS’s, because your Y capacitors are the ground itself, and also the diff caps act as Y caps…………..in other words, its far more difficult for noise to couple out of a non-isolated smps than an isolated one……this is because in an  isolated SMPS, the noise couples out of the transformer into the isolated part, and then doesn’t “see” a low impedance connection to channel it back to the input wires…therefore the noise can more easily “escape” the confines of the product and create common mode noise….this is why you should always do non-isolated SMPS where you can, in order to  better counteract common mode noise.

As discussed, when you have a non-isolated SMPS, you still have to connect your chassis to circuit ground…but you must only do this  in one place, otherwise you create a current loop…which will result in noise getting induced into the circuit…………….the single point connection will not be an electrical connection for  all the noise though…..this is because the noise is high frequency, and the distance to the single connection point will be so far (>many wavelengths)  that the connection point will look like an open circuit to certain frequencies…..as such, connecting wires from circuit to chassis ground will not result in noise getting conveniently channelled back into the circuit,  which would reduce common mode noise.

So that’s it basically, non isolated is better from a common mode noise point of view. And we have the common mode chokes  (as attached) to stop noise getting induced into the wide area current loops.

A good exercise for the student is ask them to  fit Y capacitors into  a non isolated SMPS that is fed by two wires only………….they find that  they have nowhere to connect the Y capacitors to……With a non isolated SMPS fed by two wires, you can only use a common mode choke to fight the common mode noise…..the Y caps are not needed as the contiguous copper ground effectively acts as a y capacitor….also the diff caps similarly act as Y caps  (common mode caps).

So basically, your kindly described method of post #8, would be worse from a common mode noise viewpoint.
« Last Edit: Yesterday at 01:02:02 pm by treez »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf